Soft-tissue prediction based on 3D photographs for virtual surgery planning of orthognathic surgery

Objectives: In orthognathic surgery, preoperative three-dimensional soft-tissue simulations are frequently used to determine the desired jaw displacements to enhance the facial soft tissue. This study aimed to develop and validate a deep learning-based method to predict postoperative facial soft tissue outcomes in real time for various orthognathic procedures. Methods: The study included 458 patients who underwent various orthognathic procedures. A deep learning-based method was developed based on 3D photographs for the real-time prediction of soft-tissue changes following orthognathic surgery. The developed method combined a morphable model, principal component analysis, and a feedforward neural network for predicting the effects of maxilla, mandible, and chin displacements on facial soft tissue. Prediction accuracy was evaluated by comparing the surface distance between predicted and actual postoperative soft tissues across facial regions. Results: The trained network generated postoperative facial predictions within 0.02 s, achieving a mean accuracy of 1.17 ± 0.49 mm across all surgery types. Regional accuracy ranged from 0.55 ± 0.23 mm (nose) to 1.60 ± 0.96 mm (chin). Conclusion: This study demonstrated the potential of a deep learning-based approach for real-time prediction of facial soft-tissue changes following various orthognathic procedures, achieving clinically relevant accuracy comparable to existing deep learning-based methods.